Gradient Magnitude Similarity Deviation Research Articles

Unsupervised MRI motion artifact disentanglement: introducing MAUDGAN

Objective. This study developed an unsupervised motion artifact reduction method for magnetic resonance imaging (MRI) images of patients with brain tumors. The proposed novel design uses multi-parametric multicenter contrast-enhanced T1W (ceT1W) and T2-FLAIR MRI images. Approach. The proposed framework included two generators, two discriminators, and two feature extractor networks. A 3-fold cross-validation was used to train and fine-tune the hyperparameters of the proposed model using 230 brain MRI images with tumors, which were then tested on 148 patients’ in-vivo datasets. An ablation was performed to evaluate the model’s compartments. Our model was compared with Pix2pix and CycleGAN. Six evaluation metrics were reported, including normalized mean squared error (NMSE), structural similarity index (SSIM), multi-scale-SSIM (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multi-scale gradient magnitude similarity deviation (MS-GMSD). Artifact reduction and consistency of tumor regions, image contrast, and sharpness were evaluated by three evaluators using Likert scales and compared with ANOVA and Tukey’s HSD tests. Main results. On average, our method outperforms comparative models to remove heavy motion artifacts with the lowest NMSE (18.34±5.07%) and MS-GMSD (0.07 ± 0.03) for heavy motion artifact level. Additionally, our method creates motion-free images with the highest SSIM (0.93 ± 0.04), PSNR (30.63 ± 4.96), and VIF (0.45 ± 0.05) values, along with comparable MS-SSIM (0.96 ± 0.31). Similarly, our method outperformed comparative models in removing in-vivo motion artifacts for different distortion levels except for MS- SSIM and VIF, which have comparable performance with CycleGAN. Moreover, our method had a consistent performance for different artifact levels. For the heavy level of motion artifacts, our method got the highest Likert scores of 2.82 ± 0.52, 1.88 ± 0.71, and 1.02 ± 0.14 (p-values ≪ 0.0001) for our method, CycleGAN, and Pix2pix respectively. Similar trends were also found for other motion artifact levels. Significance. Our proposed unsupervised method was demonstrated to reduce motion artifacts from the ceT1W brain images under a multi-parametric framework.

Enhancing visual clarity in rainy conditions based on single-frame filtering algorithm

Vision-based driving assistance systems has made a paradigm shift in the automotive industry, especially in auto electronics, towards vehicular technology in smart cities. This technology concentrates on safeguarding the members inside and outside the car, such as pedestrians or other vehicles on the road. This paper proposes a methodology that employs a single image/frame-based rain filter for video processing. In this paper, an innovative method of frame filtering is statistically modelled to provide a clear vision on the road during rainy weather. The frame filtering displays an improvement by considering a reference frame as the input frame and it is further exploited to confiscate the contextual and atmospheric particles using the method of frame difference. Moreover, this article suggests a hybrid algorithm that accomplishes filtering utilizing L0-gradient image smoothing and weights least square smoothing along with adaptive histogram equalization to preserve and enhance the image under intense rainy conditions. With a single frame-based filtering methodology, the proposed algorithm performs well with rainy images but loses information during video processing. The proposed method displays an improvement by 36.29 % and 3.48 % in Structural Similarity Index (SSIM) and 57.15 % and 14.47 % in Gradient Magnitude Similarity Deviation (GMSD) as compared to guided filter and bilateral filter methods.

CDRSHNET: VARIANCE-GUIDED MULTISCALE AND SELF-ATTENTION FUSION WITH HYBRID LOSS FUNCTION TO RESTORE TRAFFIC-SIGN IMAGES CAPTURED IN ADVERSE CONDITIONS

This paper proposes a CDRSHNet (CodecDirtyRainyShadowHazeNetwork) architecture with a fusion of self-attention (SA) and variance-guided multiscale attention (VGMA) mechanism to restore traffic sign images captured in challenging weather conditions including raindrops, shadows, haze, blurry images from dirty camera lenses and codec errors. The SA captures global dependencies whereas VGMA enhances the representation by emphasizing informative channels and spatial locations. To enhance the image quality hybrid loss function is proposed that combines the Gradient Magnitude Similarity Deviation (GMSD) and Charbonnier loss. The CDRSHNet is trained on a dataset of real and synthesized images. Its performance is evaluated on the average Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR) on Test RID (Real Image Dataset) and Test SID (Synthesized Image Dataset). CDRSHNet achieved an average SSIM of 0.978 and an average PSNR of 39.58 on Test RID. On Test SID, the average SSIM is 0.963, and the average PSNR is 39.46

MRI motion artifact reduction using a conditional diffusion probabilistic model (MAR-CDPM).

High-resolution magnetic resonance imaging (MRI) with excellent soft-tissue contrast is a valuable tool utilized for diagnosis and prognosis. However, MRI sequences with long acquisition time are susceptible to motion artifacts, which can adversely affect the accuracy of post-processingalgorithms. This study proposes a novel retrospective motion correction method named "motion artifact reduction using conditional diffusion probabilistic model" (MAR-CDPM). The MAR-CDPM aimed to remove motion artifacts from multicenter three-dimensional contrast-enhanced T1 magnetization-prepared rapid acquisition gradient echo (3D ceT1 MPRAGE) brain dataset with different brain tumor types. This study employed two publicly accessible MRI datasets: one containing 3D ceT1 MPRAGE and 2D T2-fluid attenuated inversion recovery (FLAIR) images from 230 patients with diverse brain tumors, and the other comprising 3D T1-weighted (T1W) MRI images of 148 healthy volunteers, which included real motion artifacts. The former was used to train and evaluate the model using the in silico data, and the latter was used to evaluate the model performance to remove real motion artifacts. A motion simulation was performed in k-space domain to generate an in silico dataset with minor, moderate, and heavy distortion levels. The diffusion process of the MAR-CDPM was then implemented in k-space to convert structure data into Gaussian noise by gradually increasing motion artifact levels. A conditional network with a Unet backbone was trained to reverse the diffusion process to convert the distorted images to structured data. The MAR-CDPM was trained in two scenarios: one conditioning on the time step of the diffusion process, and the other conditioning on both and T2-FLAIR images. The MAR-CDPM was quantitatively and qualitatively compared with supervised Unet, Unet conditioned on T2-FLAIR, CycleGAN, Pix2pix, and Pix2pix conditioned on T2-FLAIR models. To quantify the spatial distortions and the level of remaining motion artifacts after applying the models, quantitative metrics were reported including normalized mean squared error (NMSE), structural similarity index (SSIM), multiscale structural similarity index (MS-SSIM), peak signal-to-noise ratio (PSNR), visual information fidelity (VIF), and multiscale gradient magnitude similarity deviation (MS-GMSD). Tukey's Honestly Significant Difference multiple comparison test was employed to quantify the difference between the models where p-value was considered statisticallysignificant. Qualitatively, MAR-CDPM outperformed these methods in preserving soft-tissue contrast and different brain regions. It also successfully preserved tumor boundaries for heavy motion artifacts, like the supervised method. Our MAR-CDPM recovered motion-free in silico images with the highest PSNR and VIF for all distortion levels where the differences were statistically significant (p-values ). In addition, our method conditioned on t and T2-FLAIR outperformed (p-values ) other methods to remove motion artifacts from the in silico dataset in terms of NMSE, MS-SSIM, SSIM, and MS-GMSD. Moreover, our method conditioned on only t outperformed generative models (p-values ) and had comparable performances compared with the supervised model (p-values ) to remove real motionartifacts. The MAR-CDPM could successfully remove motion artifacts from 3D ceT1 MPRAGE. It is particularly beneficial for elderly who may experience involuntary movements during high-resolution MRI imaging with long acquisitiontimes.

Enhancement and labelling of OCT images

Abstract Optical Coherence Tomography (OCT) is a noninvasive technique for visualizing retinal cross-sections, assisting in diagnosing and monitoring retinal diseases. This study presents an effective OCT image noise reduction and segmentation method. Due to the high resolution of our primary image database, a visual comparison of filter performance was challenging. To address this, we utilized a high-noise image dataset from Duke University, enabling a better evaluation of noise reduction filters. Our method was compared against eight widely-used image filters, including Gaussian, Low pass,Wavelet domain filtering, Lee filter, Anisotropic diffusion, Bilateral filter, Total variational filter, and BM3D filter. Both visual and quantitative analyses were conducted using no-reference performance parameters, namely Gradient Magnitude Similarity Deviation (GMSD), Standard Deviation of Wavelet Coefficients (SDWC), Focus Measure with Tengrade Variance (FMTV), Perception-based Image Quality Evaluator (PIQE), and Visual Information Fidelity (VIF). The results demonstrated the superiority of our proposed filter in terms of noise reduction performance while maintaining the sharpness of retinal layers. Quantitative analysis revealed notable performance gains, including improvements of 63.27% to 83.24% with GMSD, consistent edge strength enhancement of 9.4% to 9.97% using SDWC, gains in image quality between 51.99% and 54.64% with FMTV, performance improvements ranging from 7.25% to 23.97% in terms of PIQE, and a substantial increase in performance varying from 16.31% to 27.69% when assessing the impact on retinal layer quality using VIF. Using the Kruskal-Wallis test, our proposed noise reduction method shows statistical significance for all quantitative parameters. Additionally, our clustering algorithm effectively separated the foreground, including retinal layers and vitreous detachment, from the background and identified an area representing the region between retinal layers where fluid accumulates. We’ve successfully achieved OCT image enhancement, along with distinct foreground and background segmentation.

Synthetic CT generation from CBCT using double-chain-CycleGAN

PurposeCone-beam CT (CBCT) has the advantage of being less expensive, lower radiation dose, less harm to patients, and higher spatial resolution. However, noticeable noise and defects, such as bone and metal artifacts, limit its clinical application in adaptive radiotherapy. To explore the potential application value of CBCT in adaptive radiotherapy, In this study, we improve the cycle-GAN's backbone network structure to generate higher quality synthetic CT (sCT) from CBCT. MethodAn auxiliary chain containing a Diversity Branch Block (DBB) module is added to CycleGAN's generator to obtain low-resolution supplementary semantic information. Moreover, an adaptive learning rate adjustment strategy (Alras) function is used to improve stability in training. Furthermore, Total Variation Loss (TV loss) is added to generator loss to improve image smoothness and reduce noise. ResultsCompared to CBCT images, the Root Mean Square Error (RMSE) dropped by 27.97 from 158.49. The Mean Absolute Error (MAE) of the sCT generated by our model improved from 43.2 to 32.05. The Peak Signal-to-Noise Ratio (PSNR) increased by 1.61 from 26.19. The Structural Similarity Index Measure (SSIM) improved from 0.948 to 0.963, and the Gradient Magnitude Similarity Deviation (GMSD) improved from 12.98 to 9.33. The generalization experiments show that our model performance is still superior to CycleGAN and respath-CycleGAN.

DoLP and AoP Synthesis from division of focal plane polarimeters using CycleGAN

Division of focal plane (DoFP) polarimeter brings significant advantages for capturing polarization information, and its unique structure enables temporal synchronization of polarization intensity measurements across scenes. While DoFP can obtain polarization information in all four directions simultaneously, it also reduces the spatial resolution of the image to 1/4 of the camera resolution. In order to solve low resolution and the cumulative error caused by the formula calculation. Using CycleGAN to construct a raw mosaic image-to-polarization properties images model, which differs from general interpolation methods and other existing deep learning methods with paired datasets and takes microgrid images as input and outputs the degree of linear polarization (DoLP) and angle of polarization (AoP) directly by learning the mapping function between unpaired the mosaic images and the polarization properties. In addition, Gradient Magnitude Similarity Deviation (GMSD) metric is applied to preserve satisfactory reconstructed image edge contour information and reduce IFOV artifacts and aliasing. The experimental results demonstrate that the quantitative performance evaluation metrics scores of our method are all higher than other methods and the reconstructed DoLP and AoP images are comparable to the ground truth imagery in terms of visual quality. The end-to-end network structure trained on unpaired datasets solves the severe limitations of paired datasets training, is more suitable for DoLP and AoP reconstructing.

CG-DRR：digital reconstructed radiograph generation algorithm based on Cycle-GAN

CG-DRR：digital reconstructed radiograph generation algorithm based on Cycle-GAN

Object Relocation Visual Tracking Based on Histogram Filter and Siamese Network in Intelligent Transportation.

Target detection and tracking algorithms are one of the key technologies in the field of autonomous driving in intelligent transportation, providing important sensing capabilities for vehicle localization and path planning. Siamese network-based trackers formulate the visual tracking mission as an image-matching process by regression and classification branches, which simplifies the network structure and improves the tracking accuracy. However, there remain many problems, as described below. (1) The lightweight neural networks decrease the feature representation ability. It is easy for the tracker to fail under the disturbing distractors (e.g., deformation and similar objects) or large changes in the viewing angle. (2) The tracker cannot adapt to variations of the object. (3) The tracker cannot reposition the object that has failed to track. To address these issues, we first propose a novel match filter arbiter based on the Euclidean distance histogram between the centers of multiple candidate objects to automatically determine whether the tracker fails. Secondly, the Hopcroft-Karp algorithm is introduced to select the winners from the dynamic template set through the backtracking process, and object relocation is achieved by comparing the Gradient Magnitude Similarity Deviation between the template and the winners. The experiments show that our method obtains better performance on several tracking benchmarks, i.e., OTB100, VOT2018, GOT-10k, and LaSOT, compared with state-of-the-art methods.

Tripartite sub-image histogram equalization for slightly low contrast gray-tone image enhancement

In this paper, a neoteric tripartite sub-image histogram equalization method is proposed to enhance slightly low contrast gray-tone images, which is a less explored area in the literature. An image is decomposed into three sub-images to preserve its mean brightness, and the histograms of the sub-images are calculated. Then, the snipping procedure is applied to each histogram to constrain the pace of contrast enhancement. Subsequently, the equalization of the three histograms is performed independently, and finally, the three equalized sub-images are composed into a single image. The proposed method offers better outcomes as compared to several common and state-of-the-art histogram equalization-based methods regarding contrast improvement, blind/reference-less image spatial quality evaluator, mean brightness preservation, peak signal-to-noise ratio, mean structural similarity, gradient magnitude similarity deviation, feature similarity, bit-plane to bit-plane similarity, and visual image quality.

Histogram modification in adaptive bi-histogram equalization for contrast enhancement on digital images

In this paper, a novel histogram modification-based bi histogram equalization (HE) approach for contrast enhancement on digital images is presented. At first, a power-logarithm transformation function is used to change the histogram of the input image. The logarithm operation reduces the input histogram’s excessive peaks, while the power function restores the histogram structure. The adjusted histogram is then separated into two sub-histograms at the threshold limit, using the input image’s minimum intensity and standard deviation. Sub-histograms are clipped based on their individual plateau limit parameters, which are based on the median value of the individual sub-histogram, to control the over-enhanced outcomes. The clipped pixels are then redistributed evenly across the histogram’s non-empty bins. Finally, the intended outcome is achieved by applying the HE procedure on the updated individual sub-histogram. Simulation results show that the proposed HE approach successfully improves the visual quality of the images. Quantitative measurements such as entropy, feature similarity index measure (FSIM), spectral residual similarity index measure (SR-SIM), gradient magnitude similarity deviation (GMSD), visual saliency-induced index (VSI) and multi scale structural similarity index measure (MS-SSIM) efficiently confirm the effectiveness of the suggested method when compared with the existing enhancement techniques. Furthermore, a comparative study of the different approaches is performed using a dataset of 300 images, demonstrating the superiority of the suggested method over the other state-of-the-art techniques.

No-reference quality assessment for neutron radiographic image based on a deep bilinear convolutional neural network

Neutron imaging (NI) has been widely employed in non-destructive investigations. Since the image quality assessment (IQA) method can be beneficial in reflecting the performance of imaging systems and image processing algorithms, we propose a proof-of-concept IQA method for the NI system based on a deep bilinear convolutional neural network (CNN) framework with two designed datasets. Due to the lack of neutron IQA database, different levels of authentic distortion induced by NI are first simulated on the natural and neutron radiographic images to generate the pre-training and fine-tuning datasets, respectively. Then, the gradient magnitude similarity deviation (GMSD) algorithm and transfer learning method are respectively employed to label the above datasets and optimize the prediction performance. Experimental results demonstrate that the proposed method can maintain good consistency with human perception in predicting the quality scores of both the authentic and enhanced neutron radiographic images.

Plant’s Leaf Analysis Using Pattern Recognition Algorithm

Plant species identification is an important area of research which is required in number of regions. Design and development of an automatic leaf based plant species identification system is a challenging task. The initial step of any coordinating/acknowledgment framework is to identify intrigue areas in the pictures and portray them. The proposed system is based on pre-processing, feature extraction and their weighted normalization and finally classification. In this paper, pattern recognition/retrieval techniques are studied by extraction of novel features in all three domains and their performance has been evaluated based on Entropy, Peak-Signal-to-Noise-Ratio (PSNR), Standard Deviation (SD), Euler Number and Gradient Magnitude Similarity Deviation. This paper presents scientific ideas for distinguishing and depicting image features.

Intensity image quality assessment based on multiscale gradient magnitude similarity deviation

Laser active imaging is widely used in many fields. The intensity image quality of laser active imaging is affected by various degradations, such as speckle effect and noise. Most existing objective image quality assessment (IQA) methods that consider only a single distorted image are not suitable for an intensity image. A multiscale full-reference intensity IQA method is presented. The proposed method is based on an improved gradient magnitude similarity deviation (GMSD) in the nonsubsampled contourlet transform (NSCT) domain. The reference and distorted images are decomposed by NSCT to emulate the multichannel structure of the human visual system. Then, the GMSD of each sub-band is computed to capture the intensity image quality. At last, the contrast sensitivity function implementation is employed in the sub-bands of the NSCT domain. All sub-bands’ GMSD is evaluated and pooled together to yield the objective quality index of a distorted intensity image. Experimental results show that the proposed method can effectively and accurately evaluate the quality of intensity images, and it is highly consistent with subjective perception.

Taxonomy of Performance Measures for Contrast Enhancement

Contrast enhancement is a primitive step in image processing application for enhancing the visual perceptibility which can be treated as the increase in gray level differences between background and foreground with suppression of noises. In literature, various methods have been proposed for contrast enhancement such as bi- histogram and two-dimensional histogram-based methods, etc. Among various methods, the suitable contrast enhancement algorithm should be chosen based on different applications. To find suitability, qualitative and quantitative analysis is also needed. The quantitative analysis can be done with the help of the various performance measures. In this paper, a complete taxonomy of performance measure for contrast enhancement has been discussed and analyzed for the CSIQ database. Among the various performance measures, it is found that Absolute Mean Brightness Error (AMBE), entropy, contrast improvement index, and gradient magnitude similarity deviation (GMSD) are better in terms of the range of operation and the wide variation of low contrast images. It is interpreted that absolute mean brightness requires less time for computation and it is widely used by the researchers for quantitative analysis of contrast enhancement algorithm. This study provides a critical analysis of the several quantitative measures that validate the effectiveness of qualitative measures and motivates to propose an alternate measure that considers the strength of various measures discussed in the literature. These measures quantify the mean brightness shifting, contrast, and entropy preservation.

GAN-Based Image Super-Resolution with a Novel Quality Loss

Single image super-resolution (SISR) has been a very attractive research topic in recent years. Breakthroughs in SISR have been achieved due to deep learning and generative adversarial networks (GANs). However, the generated image still suffers from undesired artifacts. In this paper, we propose a new method named GMGAN for SISR tasks. In this method, to generate images more in line with human vision system (HVS), we design a quality loss by integrating an image quality assessment (IQA) metric named gradient magnitude similarity deviation (GMSD). To our knowledge, it is the first time to truly integrate an IQA metric into SISR. Moreover, to overcome the instability of the original GAN, we use a variant of GANs named improved training of Wasserstein GANs (WGAN-GP). Besides GMGAN, we highlight the importance of training datasets. Experiments show that GMGAN with quality loss and WGAN-GP can generate visually appealing results and set a new state of the art. In addition, large quantity of high-quality training images with rich textures can benefit the results.

Recursively Partitioned Clipped Histogram Equalization Techniques for Preserving Image Features

In this work, we present recursively partitioned clipped histogram equalization (RPCHE) techniques, viz., recursive median partitioned clipped histogram equalization (RMDPCHE) and recursive mean partitioned clipped histogram equalization techniques for better quality images. They adopt the feature of histogram equalization (HE) in terms of simplicity, the feature of recursive histogram partition and histogram equalization to maintain low absolute mean brightness error and the feature of clipped histogram equalization in terms of control on over enhancement. In addition to these, the RPCHE methods are devoid of intensity compression, resulting in no gray level loss, and retain the total number of gray, assure uniform degree enhancement of gray to get over all image enhancement and assure no false contouring of objects. In RMDPCHE, the histogram is divided recursively with median gray level, and later these sub histograms are restricted to individual clipped threshold and finally conventional HE is applied on these clipped histograms to get over all equalized image. Experimental results show the superiority of the proposed methods over the state-of-art HE methods in terms of preserving image features with uniform degree of enhancement. They are able to achieve maximum entropy with minimum gradient magnitude similarity deviation. These ensure the objects in the processed image to have fine contours with natural enhancement.

Objective Edge Similarity Metric for denoising applications in MR images

Edge Similarity Metrics (ESMs) are necessary to objectively quantify the inadvertent blur at the edge pixels which occurs during denoising. They are helpful for evaluating edge-preserving capability of nonlinear filters. Most of the ESMs in literature, consider similarity of either strength of the edges or their direction individually. They lag in terms of concordance with subjective edge similarity ratings. An Objective Edge Similarity Metric (OESM) which considers all three attributes of edges; strength, direction and width together, is proposed in this paper. Pearson’s Correlation shown by Gradient Magnitude Similarity Deviation (GMSD), Gradient Similarity Measure (GSM), Edge Strength Similarity Index Metric (ESSIM) and OESM with Subjective Edge Similarity Score (SESS) are −0.9669 ± 0.0028, 0.9566 ± 0.0053, 0.9507 ± 0.0057 and 0.9848 ± 0.0038, respectively. OESM is able to measure the degree of edge similarity between images more efficiently than GMSD, GSM and ESSIM. It reflects the perceptual edge similarity between images more accurately than GMSD, GSM and ESSIM.

Comparison of Objective Image Quality Metrics to Expert Radiologists’ Scoring of Diagnostic Quality of MR Images

Image quality metrics (IQMs) such as root mean square error (RMSE) and structural similarity index (SSIM) are commonly used in the evaluation and optimization of accelerated magnetic resonance imaging (MRI) acquisition and reconstruction strategies. However, it is unknown how well these indices relate to a radiologist's perception of diagnostic image quality. In this study, we compare the image quality scores of five radiologists with the RMSE, SSIM, and other potentially useful IQMs: peak signal to noise ratio (PSNR) multi-scale SSIM (MSSSIM), information-weighted SSIM (IWSSIM), gradient magnitude similarity deviation (GMSD), feature similarity index (FSIM), high dynamic range visible difference predictor (HDRVDP), noise quality metric (NQM), and visual information fidelity (VIF). The comparison uses a database of MR images of the brain and abdomen that have been retrospectively degraded by noise, blurring, undersampling, motion, and wavelet compression for a total of 414 degraded images. A total of 1017 subjective scores were assigned by five radiologists. IQM performance was measured via the Spearman rank order correlation coefficient (SROCC) and statistically significant differences in the residuals of the IQM scores and radiologists' scores were tested. When considering SROCC calculated from combining scores from all radiologists across all image types, RMSE and SSIM had lower SROCC than six of the other IQMs included in the study (VIF, FSIM, NQM, GMSD, IWSSIM, and HDRVDP). In no case did SSIM have a higher SROCC or significantly smaller residuals than RMSE. These results should be considered when choosing an IQM in future imaging studies.

Performance Analysis of Fractional Redundant Wavelet Transform for Watermarking Scheme

In this paper, an algorithm for digital image watermarking which utilizes the hybridization technique is presented. The hybrid technique is formulated by combining the Redundant Wavelet Transform (RDWT) with Fractional Fourier Transform (FrFT) and Singular Value Decomposition (SVD). In this technique, watermark information is embedded in the low frequency band of Redundant Wavelet Transform. To increase the robustness, FrFT is implemented on low frequency coefficients of RDWT. Experimental results have been demonstrated on the basis of Peak Signal to Noise Ratio (PSNR), Correlation Coefficient (CC), and Gradient Magnitude Similarity Deviation (GMSD). A comparable improvement is witnessed from the results in terms of qualitative and quantitative analysis. The experimental results prove to be robust against various image processing and geometrical attacks applied on the standard test images.